Internet-based Machine Learning Platform (ML PaaS)

A Internet-based Machine Learning Platform (ML PaaS) is an 3rd-party ML platform that is a Internet-based platform.

• Context:
  • It can range from an End-to-End ML PaaS to being an ML PaaS Component.
  • It can be used to develop a Cloused-based ML System.
  • It can be associated with a Cloud-based AI Platform, such as AWS Polly text-to-speech platform.
  • …
• Example(s):
  • an End-to-End ML PaaS, such as: AWS SageMaker, GCP Vertex AI, and Azure Machine Learning.
  • …
  • Chiron System,
  • MLCapsule System.
  • …
• Counter-Example(s):
  • an On-Premise ML Platform, such as: PipelineAI Platform, MLflow Platform, Kubeflow Platform.
  • a Cloud-based ETL Platform, BI Platform..
• See: ML Model Deployment System, Machine Learning Framework, ML Development System.

---

---

References

2018a

• (Hunt et al., 2018) ⇒ Tyler Hunt, Congzheng Song, Reza Shokri, Vitaly Shmatikov, and Emmett Witchel. (2018). “Chiron: Privacy-preserving Machine Learning As a Service.” arXiv:1803.05961
  • QUOTE: ML-as-a-service platforms provide convenient APIs for users to upload their data and train an ML model. The trained model can be returned directly to the user or made available for querying through a special API. Many major cloud providers now offer this service, including Google’s Prediction API (soon to be replaced by Cloud Machine Learning Engine), Amazon ML , and Microsoft’s Azure ML .

     ML-as-a-service APIs are usually provided as black boxes. In many services, the user does not know the type of the model selected by the provider (which could depend on the user’s data and task) or the details of the training. Google’s Prediction API hides all details; users have no information about how the model is designed and trained. Amazon ML lets users choose a few hyper-parameters such as model size, regularization, and the number of training iterations. The choice of the model depends on these hyper-parameters but is invisible to the user. Model training involves stochastic gradient descent but the implementation details are hidden. Microsoft’s Azure ML provides a wide range of built-in models. Users can choose a model but have no information about the implementation details of the learning algorithm.

    Our design of Chiron preserves this separation between model design, which is proprietary to the service operator and not available to the user, and model training, which is a generic procedure of repeatedly applying the training function to batches of training data.

Google Cloud Platform: https://cloud.google.com/products/machine-learning.

Amazon machine learning: https://aws.amazon.com/machine-learning/

Microsoft Azure Machine Learning. https://azure.microsoft.com/en-us/services/machinelearning

2018b

• (Hanzlik et al., 2018) ⇒ Lucjan Hanzlik, Yang Zhang, Kathrin Grosse, Ahmed Salem, Max Augustin, Michael Backes, and Mario Fritz. (2018). “MLCapsule: Guarded Offline Deployment of Machine Learning As a Service.” arXiv:1808.00590
  • QUOTE: Machine learning as a service (MLaaS) has become increasingly popular during the past five years. Leading Internet companies, such as Google, Amazon, and Microsoft have deployed their own MLaaS. It offers a convenient way for a service provider to deploy a machine learning (ML) model and equally an instant way for a user/client to make use of the model in various applications. Such setups range from image analysis over translation to applications in the business domain. (...)

    In this paper, we propose MLCapsule, a guarded offline deployment of machine learning as a service. MLCapsule follows the popular MLaaS paradigm, but allows for clientside execution while model and computation remain secret. With MLCapsule, the service provider controls its ML model which allows for intellectual property protection and business model maintenance. Meanwhile, the user gains perfect data privacy and offline execution, as the data never leaves the client and the protocol is transparent.

---